On Wednesday, Google introduced its latest big project to join the competitive world of artificial intelligence. They launched an AI model called Gemini, aiming to rival OpenAI’s GPT models. The goal is to enhance various aspects, from Google’s apps to Android phones.
Google expressed its significant goals in the announcement, presenting Gemini as their “biggest and most powerful AI model.” They foresee a “Gemini era” where this technology will be used widely, from large corporations to everyday devices like the Google Pixel 8 Pro.
Unlike current AI models that usually handle only one kind of user prompt, focusing on either images or text, Gemini was created to be “multimodal,” according to Google. This means it can take in various types of media inputs, blending text, images, audio, video, and programming code.
Google’s Big Move, Introducing New AI Models
Google CEO Sundar Pichai mentioned in a blog post that this new generation of models marks one of the most extensive science and engineering endeavors undertaken by the company.
On Wednesday, Google shared that its exclusive AI chatbot, Bard, has received an upgrade using the Gemini model. The company plans to integrate Gemini into widely used products like Google’s search engine and the Chrome web browser, which are used by billions of people globally.
This announcement is an effort by Google to regain momentum after being seemingly caught off guard a year ago by the unexpected and widely embraced release of OpenAI’s ChatGPT. This event triggered a rush across the industry to accelerate the development of generative AI tools and sparked a global discussion about the risks and benefits of AI.
This is also an attempt to extend generative AI across the entire Google network. Google mentioned that Gemini 1.0 is available in three sizes: Nano, tailored for mobile devices and app developers; Pro, the standard model suitable for various tasks and users; and Ultra, considered the most advanced AI model ever created by Google, currently undergoing safety testing.
Google’s Progress in Cloud Computing and Gemini AI Model’s
The launch on Wednesday aimed to highlight Google’s progress in cloud computing, a crucial asset for AI developers. Google explained that it trained Gemini using a new set of potent cloud-based processors capable of collectively training large AI models nearly three times faster than the previous version.
This technology, available to Google’s cloud customers, has the potential to significantly benefit the broader AI industry by making AI training more accessible. Additionally, it could strengthen Google’s position as the third-largest player in the market for public cloud services.
However, the comparison between Google’s AI chips and those from leading chipmakers like Nvidia remains unclear.
During testing, Google’s Gemini model surpassed competing AI models in over two dozen benchmarks commonly employed by AI researchers to assess reading comprehension, mathematical capabilities, and multistep reasoning skills, according to the company.
VP Eli Collins and Safety Measures for Gemini Ultra
Eli Collins, the Vice President of Product at Google DeepMind, stated in a conference call on Tuesday, “We do see it setting new kinds of frontiers across the board.” However, he seemed to acknowledge the ongoing risk of AI models providing users with misleading results. This acknowledgment alluded to concerns expressed by civil society groups, policymakers, and AI researchers.
Collins mentioned that Google has invested significant effort into enhancing the accuracy of information in Gemini, although he acknowledged that large language models still have the potential to generate false information with high confidence, a phenomenon referred to as “hallucinating” by AI researchers.
He explained, “When we integrate these models into products like Bard, we have additional techniques to improve the accuracy of responses.”
In acknowledgment of these potential risks, Google announced on Wednesday that Gemini Ultra, the most advanced version of the model, will be gradually released to select customers, developers, partners, and safety and responsibility experts for early experimentation and feedback.
This process will precede the wider rollout to developers and enterprise customers early next year. Gemini Ultra is currently undergoing third-party safety evaluations, known as red-teaming, in line with a commitment made to the Biden administration earlier this summer.
Conclusion
Google’s recent foray into artificial intelligence with the Gemini model marks a significant leap. With ambitions to revolutionize various applications, from mobile devices to cloud computing, Google aims to lead the industry. The unveiling of Gemini Ultra, coupled with safety precautions, signals Google’s commitment to responsible AI advancement.
